The Agent Skills Directory

[PROMPT_INJECTION]: The skill instructs the agent to adopt an autonomous 'loop' behavior, prioritizing goal completion over step-by-step interaction. It explicitly defines 'Stopping Conditions' where user intervention is required for irreversible or high-risk actions (e.g., merging to main, destructive deletions, external communications, or spending credits), which acts as a safety guardrail.
[COMMAND_EXECUTION]: The instructions direct the agent to execute standard development commands such as typecheck, lint, and build, as well as end-to-end functional tests. These operations are aligned with the skill's stated purpose of facilitating software development tasks.
[INDIRECT_PROMPT_INJECTION]: The skill functions by ingesting external data such as user objectives, project specifications (tasks.md, MEMORY.md), and system logs to guide its autonomous loop.
Ingestion points: User objectives provided via arguments or conversation history; local project files including specs, tasks, and memory files.
Boundary markers: The skill does not define specific text delimiters but references an external behavioral contract in CLAUDE.md.
Capability inventory: The skill utilizes a suite of speckit tools for planning and implementation, and executes shell commands for project building and verification.
Sanitization: No explicit mention of sanitizing or escaping data read from project files before processing.

loop-sdd